Elamur Research⌕
☰
home / papers / 2013 / Improving pairwise coreference models through feature space hierarchy learning

Improving pairwise coreference models through feature space hierarchy learning

Summary

Pairwise coreference systems usually rely on one classifier to judge whether two mentions corefer. Lassalle and Denis argue this is statistically questionable, because different kinds of mention pairs (for example, those involving pronouns versus nominals) likely follow different distributions. Their solution takes a set of indicators, such as grammatical type, named entity type or sentence distance, and uses them to split pairs into disjoint classes, each with its own linear model trained by the Online Passive-Aggressive algorithm. With linear models, this split is equivalent to choosing one larger feature space. The authors organize the possible splits as hierarchies and use dynamic programming to search exactly for the hierarchy that best balances overly coarse (noisy) against overly fine (sparse) partitions. On the CoNLL-2012 English data with gold mentions, the learned hierarchies beat a single model using the same base features by a wide margin, regardless of which greedy decoder or metric is used. This offers a principled, data-driven alternative to hand-picked specialized coreference models.

Key points

  • The space of candidate hierarchies grows doubly exponentially in the number of indicators (at least 2^(2^n)), yet the search stays tractable because dynamic programming exploits the hierarchies' structure.
  • Compared with a single pairwise model, separating pairs raised the mean F1 by 6.4 to 8.3 points, depending on the greedy decoder (e.g., Closest-First).
  • The best system scored 67.19 average F1 over MUC, B³ and CEAF. This exceeds the 66.41 mean of CoNLL-2012 Shared Task participants in the gold-mentions track, although the top system reached 77.22 and other leading systems scored around 68–69.
  • The models ranked consistently across decoders: BestHierarchy > GramType > Right-Type and the single model. For instance, one effective learned ordering splits pairs first by the grammatical type of the right mention, next by that of the left mention, then by whether the two mentions occur in the same sentence, and lastly by the right mention's named entity type.
  • Full product hierarchies that were not cut back suffered from data sparsity and could score below the single model. The authors propose extending the method to anaphoricity detection, discourse and temporal relation classification, and decision-tree-based structures.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Coreference resolution

Cited by

5 papers of our library cite it (of the 140673 whose bibliographies we read); 3 of them have a page here, 2 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{lassalle2013improving,
  title = {{Improving pairwise coreference models through feature space hierarchy learning}},
  author = {Lassalle, Emmanuel and Denis, Pascal},
  booktitle = {Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2013},
  url = {https://aclanthology.org/P13-1049/}
}

References

28 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. M. Ariel. 1988. Referring and accessibility. Journal of Linguistics, pages 65–87. find
  2. A. Bagga and B. Baldwin. 1998. Algorithms for scoring coreference chains. In Proceedings of LREC 1998, pages 563–566. find our card
  3. (Bansal and Klein, 2012) show good performances of decision trees on coreference resolution. find
  4. Mohit Bansal and Dan Klein. 2012. Coreference semantics from web features. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers-Volume 1, pages 389–398. Association for Computational Linguistics. find our card
  5. Eric Bengston and Dan Roth. 2008. Understanding the value of features for coreference resolution. In Proceedings of EMNLP 2008, pages 294–303, Honolulu, Hawaii. find our card
  6. Jie Cai and Michael Strube. 2010. End-to-end coreference resolution via hypergraph partitioning. In COLING, pages 143–151. find our card
  7. Bin Chen, Jian Su, Sinno Jialin Pan, and Chew Lim Tan. 2011. A unified event coreference resolution by integrating multiple resolvers. In Proceedings of 5th International Joint Conference on Natural Language Processing, pages 102–110, Chiang Mai, Thailand, November. Asian Federation of Natural Langu… find our card
  8. Koby Crammer, Ofer Dekel, Joseph Keshet, Shai Shalev-Shwartz, and Yoram Singer. 2006. Online passive-aggressive algorithms. Journal of Machine Learning Research, 7:551–585. find
  9. Pascal Denis and Jason Baldridge. 2008. Specialized models and ranking for coreference resolution. In Proceedings of EMNLP 2008, pages 660–669, Honolulu, Hawaii. find our card
  10. Pascal Denis and Jason Baldridge. 2009. Global joint models for coreference resolution and named entity classification. Procesamiento del Lenguaje Natural, 43. find
  11. Trevor Hastie, Robert Tibshirani, and J. H. Friedman. 2001. The elements of statistical learning: data mining, inference, and prediction: with 200 full-color illustrations. New York: Springer-Verlag. find
  12. A. Kehler, D. Appelt, L. Taylor, and A. Simma. 2004. The (non)utility of predicate-argument frequencies for pronoun interpretation. In Proceedings of HLT-NAACL 2004. find
  13. M. Klenner. 2007. Enforcing coherence on coreference sets. In Proceedings of RANLP 2007. find
  14. X. Luo. 2005. On coreference resolution performance metrics. In Proceedings of HLT-NAACL 2005, pages 25–32. find our card
  15. J. F. McCarthy and W. G. Lehnert. 1995. Using decision trees for coreference resolution. In IJCAI, pages 1050–1055. find
  16. T. Morton. 2000. Coreference for NLP applications. In Proceedings of ACL 2000, Hong Kong. find
  17. V. Ng and C. Cardie. 2002. Improving machine learning approaches to coreference resolution. In Proceedings of ACL 2002, pages 104–111. find our card
  18. V. Ng. 2005. Supervised ranking for pronoun resolution: Some recent improvements. In Proceedings of AAAI 2005. find
  19. Cristina Nicolae and Gabriel Nicolae. 2006. Bestcut: A graph algorithm for coreference resolution. In EMNLP, pages 275–283. find our card
  20. Simone Paolo Ponzetto and Michael Strube. 2006. Exploiting semantic role labeling, WordNet and Wikipedia for coreference resolution. In Proceedings of the HLT 2006, pages 192–199, New York City, N.Y. find our card
  21. Sameer Pradhan, Alessandro Moschitti, Nianwen Xue, Olga Uryupina, and Yuchen Zhang. 2012. Conll-2012 shared task: Modeling multilingual unrestricted coreference in ontonotes. In Joint Conference on EMNLP and CoNLL - Shared Task, pages 1–40, Jeju Island, Korea, July. Association for Computational Lin… find our card
  22. Altaf Rahman and Vincent Ng. 2011. Narrowing the modeling gap: a cluster-ranking approach to coreference resolution. J. Artif. Int. Res., 40(1):469–521. find
  23. Recasens and Hovy. 2011. Blanc: Implementing the rand index for coreference evaluation. Natural Language Engineering, 17:485–510, 9. find
  24. W. M. Soon, H. T. Ng, and D. Lim. 2001. A machine learning approach to coreference resolution of noun phrases. Computational Linguistics, 27(4):521–544. find our card
  25. Olga Uryupina, Massimo Poesio, Claudio Giuliano, and Kateryna Tymoshenko. 2011. Disambiguation and filtering methods in using web knowledge for coreference resolution. In FLAIRS Conference. find
  26. O. Uryupina. 2004. Linguistically motivated sample selection for coreference resolution. In Proceedings of DAARC 2004, Furnas. find
  27. Yannick Versley, Alessandro Moschitti, Massimo Poesio, and Xiaofeng Yang. 2008. Coreference systems based on kernels methods. In COLING, pages 961– 968. find our card
  28. M. Vilain, J. Burger, J. Aberdeen, D. Connolly, and L. Hirschman. 1995. A model-theoretic coreference scoring scheme. In Proceedings fo the 6th Message Understanding Conference (MUC-6), pages 45–52, San Mateo, CA. Morgan Kaufmann. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/lassalle-2013-improving-pairwise-coreference-models-through/